Just as with auto safety and auto accidents, making sex safer has two effects working in opposite directions. It makes the chance that a given act of sex will result in AIDS transmission lower. But, by lowering that risk, it reduces the incentive to avoid sex entirely, to avoid sexual acts such as anal intercourse that are particularly likely to transmit AIDS, to avoid sex with people likely to give you AIDS, such as prostitutes. On theoretical grounds we have no way of knowing whether the net effect will be more AIDS or less.Todd Zywicki over at Volokh previously pointed to stronger confirmation of this offsetting effect. The Alligator, linked to by Zywicki, quotes Dr. Edward Green, Director of Harvard's AIDS Prevention Research project:
It turns out that there is evidence that, just as in the auto case, the two effects roughly cancel. That, at least, was the widely reported conclusion of a Harvard AIDS researcher who had actually looked at the data. “We have found no consistent associations between condom use and lower HIV-infection rates, which, 25 years into the pandemic, we should be seeing if this intervention was working.”
In epidemics that are population wide, where most HIV is found in the general population, for whatever reason we can't get people to use condoms consistently, and when they use them at all, that seems to have the effect of disinhibiting people's behaviours so they end up taking greater sexual risks and cancelling whatever risk reduction they have gotten from the technology they're using.
Condoms make you feel safer, so you take more risks. Always beware the offsetting behaviour.